Skip to main content
You can work with Nebius AI Cloud using a coding agent. Once you connect the agent, it can check your resources and create new ones with your confirmation. The steps on this page install three components that the agent uses:

​
Prerequisites

  1. Use a coding agent that supports local MCP servers and the Agent Skills standard.
  2. Sign up for Nebius AI Cloud.
  3. Set up billing to let the agent create resources.
  4. Install Python 3.13 or later for the local MCP server to work.
  5. Install Node.js 22.20 or later to install the skills in Cursor and other agents.

​
Steps

  1. In your coding agent, send the following setup prompt:
    The agent checks your environment, installs uv and the Nebius AI Cloud CLI if they’re missing, and adds the Nebius MCP server to its configuration. If you don’t have a CLI profile yet, the agent asks you to create one with nebius profile create.
  2. Install the skills and restart the agent:
    1. Add the Nebius plugin marketplace:
    2. Install the nebius-cloud plugin:
    3. Restart Claude Code, so that it loads the MCP server and the skills.
  3. Check that the agent has loaded the MCP server and the skills:
    In your terminal:
    • Run claude mcp list and check that the list includes the nebius server.
    • Run claude plugin list and check that nebius-cloud@nebius is enabled.
  4. Ask the agent to make a request to Nebius AI Cloud:
    The agent replies with the projects in your tenant, with their names and IDs. The agent may ask for permission to run a nebius command. Check that the command only lists projects, then allow it.

​
What’s next

Ask the agent what you can build on Nebius AI Cloud, or give it a task. For example:
To limit what the agent can do, see the security recommendations for the agent skills and the Nebius MCP server.

​
Troubleshooting

  1. Restart the agent. Agents load MCP servers when they start.
  2. Check the configuration as described in step 3 of Steps. If the nebius server isn’t listed, send the setup prompt again.
  3. If the server is listed but doesn’t start, check that uvx --version and python3 --version work in your terminal and that Python is 3.13 or later. For other causes, see the MCP documentation of your agent, for example, Claude Code, Codex, or Cursor.
  1. Check that the skills are loaded as described in step 3 of Steps. If they aren’t, install them again.
  2. Update the Nebius AI Cloud CLI to the latest version by running nebius update.
  3. Update the skills to the latest version.
If a command fails with the PERMISSION_DENIED error, the account in your CLI profile doesn’t have the IAM role that the action requires. Ask an administrator of your tenant to add your account to a group that has the role. For details, see Roles and Managing custom groups.If the agent uses a CLI profile with limited roles, run the command yourself or switch to a profile that has the role.
To create resources, your account needs billing details. Set up billing in the web console, then ask the agent to try again.
If a command fails with the RESOURCE_EXHAUSTED error, a quota in your tenant is used up. Ask the agent which quota it reached, then delete the resources that you no longer need or request a quota increase.
For a first test, the agent may suggest a more powerful GPU platform than you need. Ask it to estimate the price before it creates anything, and to use a preemptible VM or a cheaper GPU platform that is available in the region of your project.
  • If the agent can’t find the CLI, install it or send the setup prompt again.
  • If the agent works with the wrong tenant or project, ask it to run nebius profile current. The CLI uses the active profile, unless the --profile parameter or the NEBIUS_PROFILE environment variable sets another one. Check that neither is set in the agent’s environment or in its MCP server configuration, or switch the active profile with nebius profile activate <profile_name>. For details, see How to configure the CLI.

​
How to disconnect your agent

  1. Remove the MCP server and the skills from the agent:
  2. If you created credentials only for the agent, such as a CLI profile or a service account, delete them. Keep the CLI profiles that you use with other tools.